Data & BI

Sales forecasting methods for SMEs: where to start

Simple sales forecasting methods an SME can explain: weighted pipeline, moving averages, seasonality, the data you need, and how to measure forecast accuracy.

Every month, someone in your business guesses next quarter’s sales. The buyer orders stock, the operations lead plans shifts, the finance lead checks the cash, and each uses a different number, often a figure from a sales meeting that nobody can explain afterwards.

You do not need a data science team to do better. The most useful sales forecasting methods for an SME are simple ones you can explain in a sentence: a weighted pipeline, a moving average, a seasonal adjustment. This guide shows where to start: the question to settle first, the data you need, four methods, a worked example, how to measure forecast accuracy, and how to use the result for stock, staffing and cash.

Before you forecast: decide the question

A forecast is a tool for a decision, so start from the decision. Three choices shape everything that follows.

  • The horizon. Stock orders placed three months ahead need a three-month forecast. Weekly rotas need a weekly one.
  • The level of detail. Total revenue, product families, individual products, regions, sales channels. The finer the level, the noisier the numbers. Most SMEs start with monthly totals and a handful of product families.
  • The users. Who reads the forecast, and what will they do with it? If nobody changes a decision because of it, it is a report, not a forecast.

Agree on one rule from the start: a forecast is not a target. The target is what you want to sell; the forecast is what you expect to sell. Mix the two and the number gets pushed up to please, then missed.

The data you need to forecast sales

You probably have more than you think, spread across several tools.

  • Sales history: invoices or orders by date, customer, product and amount, ideally over two or three years so that seasons repeat. Your accounting software or ERP (Enterprise Resource Planning) system usually holds it.
  • The pipeline: open opportunities with their stage, value and expected closing date, normally in a CRM (Customer Relationship Management) system. If the real pipeline lives in spreadsheets and inboxes, that is one of the signs you need a CRM.
  • Calendar effects: holidays, trade fairs, promotions, price changes, a large one-off order. Write them down as you go: a year later, nobody remembers why March jumped.

Then clean it: remove duplicates, separate credit notes, mark one-off orders, and code product families the same way every year. When the data sits in several systems, our guide to the data warehouse explains when one is worth it and when a simpler set-up will do.

At Innova Home in Morocco, our founder, Anass Chabbab, in his previous roles, led the move from paper and Excel to a single order-to-invoice tool, which saved more than 50% of the time spent on each order. A single tool like that is also what makes order history usable later on.

Simple sales forecasting methods you can explain

Start with the simplest method that answers your question, then add complexity only if it measurably reduces the error.

Weighted pipeline

For businesses that sell through deals (business-to-business, or B2B, services, projects, equipment), multiply each open opportunity by the probability of winning it at its stage, then add the results. Use your own historical win rates by stage, taken from your closed deals, not the default percentages a CRM ships with. This method suits a horizon of one to three months, because beyond that most future sales are not in the pipeline yet.

Moving averages

For repeat sales (consumables, subscriptions, retail), average the last three, six or twelve months and use that as next month’s forecast. A short window reacts quickly but jumps with noise; a long one is stable but slow. It is crude, and everyone understands it.

Seasonality

If December is always busy and August always quiet, a plain average will be wrong twice a year. Work out a seasonal index for each month: that month’s sales divided by the average month, over several years. Then multiply your baseline by the index. The same idea supports demand forecasting for stock: forecast units by product family, not just revenue.

Basic statistical models

Next come simple statistical models. Exponential smoothing gives more weight to recent months and can handle trend and seasonality together. A linear regression can relate sales to a driver you know in advance, such as the number of active customers or a promotion calendar. Spreadsheets and most BI (Business Intelligence) tools can run both.

Worked example: a quarter forecast

Illustrative example (fictitious figures). A distributor sells to trade customers through quotes, plus repeat orders from existing accounts. It forecasts the next quarter in two parts.

Part 1, new business from the pipeline, using win rates from its last two years of closed deals:

Stage Open value Historical win rate Weighted value
First meeting 200,000 10% 20,000
Proposal sent 120,000 30% 36,000
Negotiation 60,000 60% 36,000
Total 380,000 92,000

Part 2, repeat orders: the six-month moving average is 100,000 a month. Over the past three years, this quarter has run 10% above an average quarter, so the seasonal index is 1.1. Repeat orders: 100,000 × 3 × 1.1 = 330,000.

Forecast for the quarter: 92,000 + 330,000 = 422,000. The team also writes a low case (pipeline win rates halved, no seasonal uplift: 346,000) and a high case, so that stock and cash plans have a range rather than a single number.

How to measure forecast accuracy

Keep each forecast as it was made, then compare it with what actually happened.

  • Forecast error per period: actual minus forecast, in value and in percentage.
  • Average percentage error (MAPE, Mean Absolute Percentage Error): the average of the errors without their sign, as a share of actual sales. It misleads on very small figures, so use it on totals or product families.
  • Bias: the average error with its sign. Over-forecasting month after month means the method, or the people feeding it, are optimistic.
  • A naive benchmark: compare your method with “same as last year” or “same as last month”. If a sophisticated method cannot beat the naive forecast, keep the naive one.

Measure at the horizon you decided on: if stock is ordered three months ahead, judge the forecast made three months before, not the one updated last week. Put forecast against actual on the same monthly dashboard as your other indicators; our guide to an executive dashboard shows how to choose the few that matter.

Using forecasts for stock, staffing and cash

The forecast earns its keep when it changes decisions.

  • Stock: order against the forecast by product family, plus a safety margin sized on your usual error.
  • Staffing: turn forecast volumes into hours of work (orders to pick, files to process, visits to make) and plan rotas and temporary help from the base case, with the high case as a trigger.
  • Cash: feed the forecast into your cash-flow plan with your real payment delays. A strong quarter in sales can still be a tight quarter in cash if customers pay at 60 days.

Review once a month: compare, adjust, write down why. A thirty-minute monthly meeting with sales, operations and finance around one shared number is often the biggest gain of the whole exercise. Setting up that routine and the data behind it is the kind of work our data and BI offer covers.

Frequently asked questions

How do I forecast sales with no sales history?

Start from your pipeline and from bottom-up assumptions you can check: number of prospects contacted, conversion rate, average order value. Replace each assumption with your own figures as soon as a few months of sales exist. Record each forecast from day one, so that you can measure its error and learn which assumptions were wrong.

What is a good forecast accuracy?

There is no universal figure: it depends on your business, your level of detail and your horizon. The useful question is whether your error is falling over time and whether your method beats a naive “same as last year” forecast. Track the trend of your error and its bias rather than chasing an industry number.

Do we need AI to forecast sales?

Usually not to start. Weighted pipelines, moving averages and seasonal indexes cover most SME needs and can be explained to everyone who uses them. Machine learning becomes worth testing when you forecast hundreds of products across many sites, with years of clean data and outside drivers such as weather or prices, and only if it beats the simpler methods on the same error measure.

In short

Start from the decision the forecast serves, then gather and clean your sales history and pipeline. Use the simplest methods first: weighted pipeline for deals, moving averages and seasonal indexes for repeat sales. Measure forecast error, bias and the naive benchmark every month, and use ranges to plan stock, staffing and cash.

Does every department in your business work from a different sales number? Book a free 30-minute discovery call: we look at your data and your planning decisions together, and see where a simple forecast would help first.

Related servicesData & BICRM

Have a project in mind?

Let’s talk about it on a 30-minute discovery call, free and with no obligation.

Book a discovery call

Further reading

All articles